di-sukharev

Iterative code review workflow for task-scoped active git changes using fresh independent reviewer agents without receiving the orchestrator's conversation history. Test whether another engineer can understand and safely maintain the change while reducing correctness, security, and regression risk. Use when the user invokes `/loop-code-review`, asks for a looped sub-agent code review, or asks the coding agent to keep reviewing and fixing current-task changes until validation passes and an independent reviewer either rates the result at least 9.5 out of 10 or reports no actionable comments.

Map the minimal evidence-backed implementation, coupling, test, and repository-constraint surface for a non-trivial code task by delegating one fresh repository scout instructed to remain read-only to gpt-5.6-terra. Use when explicitly invoked with `$code-scout`, or before implementing, debugging, reviewing, or planning a cross-file or cross-layer change whose owning files, symbols, entry points, or closest tests are not already known. Skip for documentation-only work, exact file-and-symbol edits, obvious local copy or style changes, tasks that need no repository search, repeated discovery for an unchanged task scope, or when the user forbids delegation.

Behavior-preserving refactoring workflow for maintainability and developer experience across frontend and backend code, run as an iterative loop with fresh independent reviewer agents until validation passes and the result scores at least 9.5/10 or has no actionable findings. Use when the user invokes `/refactoring`, asks to refactor or clean up code, improve DX or maintainability, polish a task's implementation before finishing, or enforce component-owned styling in UI code. Includes task-scoped discovery, friction mapping, challenge review, characterization tests, minimal implementation, an independent review loop, and a "no changes needed" outcome.